--- title: Diffusion Distance Based Loss keywords: fastai sidebar: home_sidebar nb_path: "05c05a Diffusion map embedding with an autoencoder.ipynb" ---
{% raw %}
{% endraw %}

It will be helpful to compare the two different distance matrices, so here's a visualization function for that:

{% raw %}

compare_distance_matrices[source]

compare_distance_matrices(embedded_points, FE, **kwargs)

{% endraw %} {% raw %}
{% endraw %} {% raw %}

class DiffusionDistanceFlowEmbedder[source]

DiffusionDistanceFlowEmbedder(X, flows, labels, device=device(type='cpu')) :: FETrainer

{% endraw %} {% raw %}
{% endraw %} {% raw %} 
from directed_graphs.datasets import directed_swiss_roll_sklearn, plot_directed_3d
X, flow, labels = directed_swiss_roll_sklearn(num_nodes=2000)
{% endraw %} {% raw %} 
X = torch.tensor(X).float().to(device)
flow = torch.tensor(flow).float().to(device)
X = X.float().to(device)
flow = flow.float().to(device)
{% endraw %} {% raw %} 
DODO_FET = DiffusionDistanceFlowEmbedder(X, flow, labels = labels, device = device)

/gpfs/loomis/project/krishnaswamy_smita/kjm76/directed_graphs/directed_graphs/utils.py:12: UserWarning: Sparsifying input to bsr_array
  warnings.warn("Sparsifying input to bsr_array")

using sigma =  2.1068635
[1.0000005 +0.j 0.98382694+0.j 0.9766961 +0.j 0.97409946+0.j
 0.95284826+0.j]

/gpfs/loomis/project/krishnaswamy_smita/kjm76/conda_envs/flowembed/lib/python3.9/site-packages/matplotlib/transforms.py:775: ComplexWarning: Casting complex values to real discards the imaginary part
  points = np.asarray(points, float)
{% endraw %} {% raw %} 
DODO_FET.fit()

  0%|          | 0/100 [00:00<?, ?it/s]
  1%|          | 1/100 [00:01<03:02,  1.84s/it]
  2%|▏         | 2/100 [00:03<02:53,  1.77s/it]
  3%|▎         | 3/100 [00:05<02:49,  1.75s/it]
  4%|▍         | 4/100 [00:07<02:49,  1.76s/it]
  5%|▌         | 5/100 [00:08<02:45,  1.75s/it]
  6%|▌         | 6/100 [00:10<02:43,  1.74s/it]
  7%|▋         | 7/100 [00:12<02:47,  1.80s/it]
  8%|▊         | 8/100 [00:14<02:42,  1.77s/it]
  9%|▉         | 9/100 [00:15<02:38,  1.74s/it]
 10%|█         | 10/100 [00:17<02:34,  1.72s/it]
 11%|█         | 11/100 [00:19<02:31,  1.70s/it]
 12%|█▏        | 12/100 [00:20<02:29,  1.69s/it]
 13%|█▎        | 13/100 [00:22<02:26,  1.69s/it]
 14%|█▍        | 14/100 [00:24<02:24,  1.68s/it]
 15%|█▌        | 15/100 [00:25<02:22,  1.68s/it]
 16%|█▌        | 16/100 [00:27<02:21,  1.68s/it]
 17%|█▋        | 17/100 [00:29<02:24,  1.75s/it]
 18%|█▊        | 18/100 [00:31<02:21,  1.73s/it]
 19%|█▉        | 19/100 [00:32<02:18,  1.71s/it]
 20%|██        | 20/100 [00:34<02:15,  1.70s/it]
 21%|██        | 21/100 [00:36<02:13,  1.69s/it]
 22%|██▏       | 22/100 [00:37<02:11,  1.68s/it]
 23%|██▎       | 23/100 [00:39<02:08,  1.67s/it]
 24%|██▍       | 24/100 [00:41<02:07,  1.67s/it]
 25%|██▌       | 25/100 [00:42<02:05,  1.67s/it]
 26%|██▌       | 26/100 [00:44<02:03,  1.67s/it]
 27%|██▋       | 27/100 [00:46<02:06,  1.73s/it]
 28%|██▊       | 28/100 [00:48<02:09,  1.79s/it]
 29%|██▉       | 29/100 [00:49<02:04,  1.75s/it]
 30%|███       | 30/100 [00:51<02:00,  1.73s/it]
 31%|███       | 31/100 [00:53<01:57,  1.71s/it]
 32%|███▏      | 32/100 [00:54<01:55,  1.70s/it]
 33%|███▎      | 33/100 [00:56<01:53,  1.69s/it]
 34%|███▍      | 34/100 [00:58<01:51,  1.68s/it]
 35%|███▌      | 35/100 [01:00<01:53,  1.75s/it]
 36%|███▌      | 36/100 [01:01<01:50,  1.73s/it]
 37%|███▋      | 37/100 [01:03<01:52,  1.78s/it]
 38%|███▊      | 38/100 [01:05<01:50,  1.78s/it]
 39%|███▉      | 39/100 [01:08<02:13,  2.19s/it]
 40%|████      | 40/100 [01:11<02:28,  2.47s/it]
 41%|████      | 41/100 [01:14<02:37,  2.67s/it]
 42%|████▏     | 42/100 [01:18<02:42,  2.81s/it]
 43%|████▎     | 43/100 [01:21<02:45,  2.90s/it]
 44%|████▍     | 44/100 [01:24<02:45,  2.96s/it]
 45%|████▌     | 45/100 [01:26<02:26,  2.66s/it]
 46%|████▌     | 46/100 [01:27<02:07,  2.36s/it]
 47%|████▋     | 47/100 [01:29<01:58,  2.23s/it]
 48%|████▊     | 48/100 [01:31<01:47,  2.07s/it]
 49%|████▉     | 49/100 [01:33<01:40,  1.96s/it]
 50%|█████     | 50/100 [01:34<01:33,  1.88s/it]
 51%|█████     | 51/100 [01:36<01:28,  1.82s/it]
 52%|█████▏    | 52/100 [01:38<01:25,  1.77s/it]
 53%|█████▎    | 53/100 [01:39<01:22,  1.76s/it]
 54%|█████▍    | 54/100 [01:41<01:19,  1.74s/it]
 55%|█████▌    | 55/100 [01:43<01:17,  1.72s/it]
 56%|█████▌    | 56/100 [01:44<01:14,  1.70s/it]
 57%|█████▋    | 57/100 [01:46<01:16,  1.78s/it]
 58%|█████▊    | 58/100 [01:48<01:13,  1.76s/it]
 59%|█████▉    | 59/100 [01:50<01:11,  1.73s/it]
 60%|██████    | 60/100 [01:52<01:08,  1.72s/it]
 61%|██████    | 61/100 [01:53<01:06,  1.70s/it]
 62%|██████▏   | 62/100 [01:55<01:04,  1.69s/it]
 63%|██████▎   | 63/100 [01:57<01:02,  1.69s/it]
 64%|██████▍   | 64/100 [01:58<01:00,  1.68s/it]
 65%|██████▌   | 65/100 [02:00<00:58,  1.68s/it]
 66%|██████▌   | 66/100 [02:02<00:57,  1.68s/it]
 67%|██████▋   | 67/100 [02:03<00:57,  1.75s/it]
 68%|██████▊   | 68/100 [02:05<00:55,  1.73s/it]
 69%|██████▉   | 69/100 [02:07<00:53,  1.72s/it]
 70%|███████   | 70/100 [02:09<00:51,  1.71s/it]
 71%|███████   | 71/100 [02:10<00:49,  1.71s/it]
 72%|███████▏  | 72/100 [02:12<00:47,  1.71s/it]
 73%|███████▎  | 73/100 [02:14<00:46,  1.72s/it]
 74%|███████▍  | 74/100 [02:15<00:44,  1.71s/it]
 75%|███████▌  | 75/100 [02:17<00:42,  1.70s/it]
 76%|███████▌  | 76/100 [02:19<00:42,  1.79s/it]
 77%|███████▋  | 77/100 [02:21<00:40,  1.77s/it]
 78%|███████▊  | 78/100 [02:22<00:38,  1.75s/it]
 79%|███████▉  | 79/100 [02:24<00:36,  1.73s/it]
 80%|████████  | 80/100 [02:26<00:34,  1.72s/it]
 81%|████████  | 81/100 [02:28<00:32,  1.71s/it]
 82%|████████▏ | 82/100 [02:29<00:30,  1.71s/it]
 83%|████████▎ | 83/100 [02:31<00:28,  1.70s/it]
 84%|████████▍ | 84/100 [02:33<00:27,  1.70s/it]
 85%|████████▌ | 85/100 [02:35<00:26,  1.76s/it]
 86%|████████▌ | 86/100 [02:36<00:24,  1.74s/it]
 87%|████████▋ | 87/100 [02:38<00:22,  1.73s/it]
 88%|████████▊ | 88/100 [02:40<00:20,  1.71s/it]
 89%|████████▉ | 89/100 [02:41<00:18,  1.70s/it]
 90%|█████████ | 90/100 [02:43<00:16,  1.70s/it]
 91%|█████████ | 91/100 [02:45<00:15,  1.69s/it]
 92%|█████████▏| 92/100 [02:46<00:13,  1.69s/it]
 93%|█████████▎| 93/100 [02:48<00:12,  1.75s/it]
 94%|█████████▍| 94/100 [02:50<00:11,  1.84s/it]
 95%|█████████▌| 95/100 [02:53<00:10,  2.09s/it]
 96%|█████████▌| 96/100 [02:55<00:07,  1.97s/it]
 97%|█████████▋| 97/100 [02:56<00:05,  1.88s/it]
 98%|█████████▊| 98/100 [02:58<00:03,  1.82s/it]
 99%|█████████▉| 99/100 [03:00<00:01,  1.79s/it]
100%|██████████| 100/100 [03:01<00:00,  1.82s/it]
{% endraw %} {% raw %} 
DODO_FET.visualize_embedding()
{% endraw %} {% raw %} 
DODO_FET.training_gif()
{% endraw %} {% raw %} 
DODO_FET.FE.precomputed_distances

tensor([[0.0000e+00, 4.1507e-02, 1.2181e-01,  ..., 1.2438e+00, 1.2464e+00,
         1.2330e+00],
        [4.1507e-02, 9.7201e-04, 1.6310e-01,  ..., 1.2434e+00, 1.2488e+00,
         1.2310e+00],
        [1.2181e-01, 1.6310e-01, 0.0000e+00,  ..., 1.2450e+00, 1.2394e+00,
         1.2387e+00],
        ...,
        [1.2438e+00, 1.2434e+00, 1.2450e+00,  ..., 0.0000e+00, 8.6444e-02,
         4.8386e-02],
        [1.2464e+00, 1.2488e+00, 1.2394e+00,  ..., 8.6444e-02, 0.0000e+00,
         1.3239e-01],
        [1.2330e+00, 1.2310e+00, 1.2387e+00,  ..., 4.8386e-02, 1.3239e-01,
         0.0000e+00]], device='cuda:0')
{% endraw %} {% raw %} 
DODO_FET.visualize_loss()
{% endraw %} {% raw %} 
plt.scatter(DODO_FET.FE.diff_coords[:,0].cpu().numpy(),DODO_FET.FE.diff_coords[:,1].cpu().numpy(),c=labels)
{% endraw %} {% raw %} 
torch.max(DODO_FET.FE.precomputed_distances)
{% endraw %} {% raw %} 
np.amax(DODO_FET.FE.diff_coords.cpu().numpy())
{% endraw %} {% raw %} 
DODO_FET.FE.diff_coords.cpu().numpy()
{% endraw %} {% raw %} 
DODO_FET.losses['diffusion']
{% endraw %} {% raw %} 
import glob
import ipywidgets as widgets
from PIL import Image
import os
import ipywidgets as widgets
import base64
frames = [Image.open(image) for image in glob.glob(f"visualizations/{BOBO_FET.timestamp}/*.jpg")]
frame_one = frames[0]
frame_one.save(f"{BOBO_FET.title}.gif", format="GIF", append_images=frames,
           save_all=True, duration=300, loop=0)
# display in jupyter notebook
b64 = base64.b64encode(open(f"{BOBO_FET.title}.gif",'rb').read()).decode('ascii')
display(widgets.HTML(f'<img src="data:image/gif;base64,{b64}" />'))
{% endraw %}